-
Notifications
You must be signed in to change notification settings - Fork 1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
generic: sycl: Inner Product FWD #2214
Conversation
@AD2605 you should probably update https://github.com/oneapi-src/oneDNN/blob/main/src/gpu/generic/sycl/README.md to mention inner product is supported :) |
@sgeor255 , updated the documentation |
3bd9d00
to
c5d66f1
Compare
a3b91d6
to
9c63d68
Compare
@oneapi-src/onednn-arch @oneapi-src/onednn-doc gentle nudge for reviewing the PR |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One minor comment added. Please incorporate as you see fit.
5899fe4
to
6989cb2
Compare
&& attr()->has_default_values(sm::scales_runtime | ||
| sm::zero_points_runtime | sm::post_ops | ||
| sm::dropout | sm::scales_runtime_data_type | ||
| sm::zero_points_runtime_data_type) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why not defer attribute and datatype checking to init_matmul?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Done in #2248, thanks.
| sm::zero_points_runtime_data_type) | ||
&& memory_desc_wrapper(src_md()).is_plain() | ||
&& memory_desc_wrapper(dst_md()) | ||
.is_plain(); // Blocked memory formats are not supported |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think is_plain is not enough to flatten the dimensions.
Here you need to check:
- that strides are dense (see is_dense).
- that the order of flattened dimensions for src and weights match.
- there is a missing check for weights_md()
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That's a good point, thanks. I had to create a new PR (#2248) to address the comments as I don't have permissions to push to this one. I think we can close this one.
Description
Implements the forward pass of the inner product for the generic sycl backend with support for different post-ops and data types